Rationality of normalization of arrays, when the 'new' mean is out of 95% cI of initial expression level?

Question

[english.server]

Hi all!

I have a replicated gene expression experiment. When I normalize expression values, for some of the probes, the 'mean of normalized values' is not within 95% cI of mean (obtained from un-normalized values). What does that statistically mean? (Before recoursing to normalization for boutique arrays).

Thank you in advance.

[Jean-Karim Heriche]

Why would you expect the normalized mean to be in the 95% confidence interval of the unnormalized one ?

[english.server]

Intuition (?). I don't think I would trust a 'blind' algorithm pushing a probe data (for 2 or 3 replicates) away from their mean. However, I have not 'seen' anything like what I said. Good question you made, however.